Struct CoClusterSelectResult 

#[non_exhaustive]
pub struct CoClusterSelectResult {
    pub best: CoClusterResult,
    pub best_k: usize,
    pub best_l: usize,
    pub grid_scores: Vec<(usize, usize, f64, usize, f64)>,
    pub slope_estimate: f64,
    pub penalty_rate: f64,
}

Result of co_cluster_select: the slope-heuristic-selected (K, L) fit together with full grid diagnostics.

Grid diagnostics

grid_scores contains one entry per (K, L) pair in the sweep: (K, L, log_likelihood, model_dim, penalised_score).

penalised_score = log_likelihood − penalty_rate × model_dim. In fallback branches (single cell, flat slope, small grid) penalised_score = log_likelihood.

Slope heuristic calibration

The Birgé–Massart penalty is estimated by OLS over the large-model (top-50% by dimension) region of the fitted grid. This is a data-driven heuristic: it works best when the grid spans a range of model dimensions and the data is well-separated enough for the log-likelihood to grow linearly with dimension in the overparameterised region. On poorly separated data the slope may be noisy and the selection may land at a boundary. Inspect grid_scores to audit the selection.

Divergence from R funHDDC

The slope calibration here uses OLS over the top-50% by model dimension (the “linear region” heuristic of Baudry, Maugis & Michel 2012). R’s funHDDC uses a slightly different calibration based on the full grid. The selected model may differ on small grids.

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

best: CoClusterResult

The selected (K*, L*) co-clustering result.

best_k: usize

Selected number of row clusters K*.

best_l: usize

Selected number of column clusters L*.

grid_scores: Vec<(usize, usize, f64, usize, f64)>

All grid fits: (K, L, log_likelihood, model_dim, penalised_score).

penalised_score = log_likelihood − penalty_rate * model_dim. In fallback branches (< 4 grid points, flat slope, etc.) penalised_score = log_likelihood.

slope_estimate: f64

OLS slope estimated from the top-50% of fits by model dimension. Zero when the grid is too small or the slope heuristic fell back to max-LL.

penalty_rate: f64

Penalty rate applied per model dimension: 2 × |slope_estimate|. Zero in fallback branches.

Trait Implementations

impl Clone for CoClusterSelectResult

fn clone(&self) -> CoClusterSelectResult

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for CoClusterSelectResult

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.